Inhabitant agent framework completes most long-horizon household tasks
Original title: Inhabitant: An Agent Framework for Long-Horizon Human-Scene Interaction with Spatial Control and Quantitative Feedback
On the 75-task Inhabit benchmark, Claude Opus 5.5 finishes 98.7% of whole-body household interaction sequences.
Inhabitant is a training-free agent framework for long-horizon human-scene interaction in 3D scenes. The agent specifies each motion with positions, paths and poses, measures deviation from intent, and revises the control specification. The authors also introduce Inhabit, 75 household tasks across 31 furniture layouts, each chaining one to three goals on movable objects and executed by the whole body excluding fingers. Reported completion is 98.7% with Claude Opus 5.5, 93.3% with Claude Sonnet 5.5 and 82.7% with Gemini 4 Argon.
- Inhabit covers 75 household tasks in 31 apartment furniture layouts.
- Tasks chain one to three goals and exclude finger motion.
- Claude Sonnet 5.5 completes 93.3% and Gemini 4 Argon 82.7%.
- Ablations say removing spatial control causes most tasks to fail.
- Without quantitative feedback, executed motion is less precise.
- Products
- InhabitantInhabit
From the original article
1 min read at huggingface.coRecent work on human-scene interaction (HSI) generation moves from single interactions to long-horizon interaction, where a language-model agent plans and executes a sequence of interactions in a 3D scene under one instruction. However, existing agents struggle in challenging everyday scenarios (e. g.
By Yao Zhang, Zhuchenyang Liu, Yu Xiao · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.